attention color whizes: non-typical sRGB/RGB/ProPhoto question

[Iliah]

> There is no such thing as abscol when the target profile is matrix based

Just a couple of things. We start in raw and it is assigning colour profile operation, typical for input devices. Only after that comes conversion. On top of that not all converters use "current CMMs".

[tho_mas]

> There is no such thing as abscol when the target profile is matrix based

Just a couple of things. We start in raw and it is assigning colour profile operation, typical for input devices. Only after that comes conversion. On top of that not all converters use "current CMMs".

Sure, but the color space initially assigned to the RAW file is a large color space the respctive RAW converter is based on, then a "camera profile" (or correction matrix) is assigned and first then it is converted to a working space like AdobeRGB or ProPhotoRGB.

Some converters have a built in CMM... but please tell me one converter that can convert from the input profile to a matrix based working space with any other RI than relcol. Talking about the relation of AdobeRGB and ProPhoto in color conversions we always talk about relcol.

[Iliah]

> the color space initially assigned to the RAW file is a large color space
It is not a good idea to assign a profile to a raw file. RGGB raw is not RGB in the meaning only one component per pixel is known, and G1 is not always filtered the same way as G2. I think saying "large" one should somehow quantify it, at least by comparison to some other known colour space.

> the respctive RAW converter is based on
I'm not sure what you mean saying a raw converter is "based" on some colour space.

> then a "camera profile" (or correction matrix) is assigned and first then it is converted to a working space like AdobeRGB or ProPhotoRGB.
Not sure I understand the above - at least one of the steps you are describing is unnecessary.

> Some converters have a built in CMM... but please tell me one converter that can convert from the input profile to a matrix based working space with any other RI than relcol.
I know a couple.

> Talking about the relation of AdobeRGB and ProPhoto in color conversions we always talk about relcol.
No.

[tho_mas]

> the color space initially assigned to the RAW file is a large color space
It is not a good idea to assign a profile to a raw file. RGGB raw is not RGB in the meaning only one component per pixel is known, and G1 is not always filtered the same way as G2. I think saying "large" one should somehow quantify it, at least by comparison to some other known colour space.

or converted. Not essential here as the target is in any case not a working space.

> the respctive RAW converter is based on
I'm not sure what you mean saying a raw converter is "based" on some colour space.

for instance ACR/LR are based on ProPhoto primaries with a linear Gamma.

> then a "camera profile" (or correction matrix) is assigned and first then it is converted to a working space like AdobeRGB or ProPhotoRGB.
Not sure I understand the above - at least one of the steps you are describing is unnecessary.

re ACR/LR: RAW->ProPhoto linear->"camera profile" (internal with "floating" white point)-> 1 of 4 possible working spaces (output spaces)
re Capture One: RAW->internal color space-> camera profile based on characterization data and tweaked to provide a gamma 1.8 TRC and to provide a certain look (think of tablebased "camera working spaces")-> output color space
(as C1's "camera profiles" are ICC based you can also embed the "camera profile" on output).

> Some converters have a built in CMM... but please tell me one converter that can convert from the input profile to a matrix based working space with any other RI than relcol.
I know a couple.

oh, that's nice for you. thanks for the valuable contribution. I've learnt a lot!

> Talking about the relation of AdobeRGB and ProPhoto in color conversions we always talk about relcol.
No.

thanks again for the comprehensive reply


_______________________________


edit: I stand corrected re abscol.
With Adobe's CMM (i.e. ACE) the only RI available is relcol.
Apple CMM does take the different white points into account as does C1's and RAW Developper's CMM.
I had this article (#21) in mind... but it doesn't refer to abscol -> http://www.colorwiki.com/wiki/Color_Management_Myths_21-25
Sorry!

[Peter_DL]

edit: I stand corrected re abscol.
With Adobe's CMM (i.e. ACE) the only RI available is relcol.
Apple CMM does take the different white points into account as does C1's and RAW Developper's CMM.

AbsCol between matrix spaces was definitively possible with Adobe’s color engine in former Photoshop CS.
It does not seem to be possible anymore with Adobe’s color engine in CS 4.  As suggested above, changing to Microsoft’s ICM does the trick– for the purpose of testing.
 
 

Re. > We start in raw and it is assigning colour profile operation, typical for input devices. Only after that comes conversion.

... but please tell me one converter that can convert from the input profile to a matrix based working space with any other RI than relcol.

Following details on ACR’s matrix profiles were given by Thomas Knoll (Feb. 2004, Rob Galbraith Forum).
It predates the more complex dng profiles which we have now.

Anyway, I’d be surprised if the initial conversion of demosaiced "RAW" RGB data to linear gamma ProPhoto RGB (via such matrices) would include a RelCol-type chromatic adaptation of the white point. It would seem to counteract the sense of the Temp. and Tint slider. Of course, I can’t know.

Peter

----

Quote: >> Camera Raw has 3 by 3 matrix values built in for all its supported cameras (actually two sets, one for 2850K and one for D65, and uses interpolation between them). But these values are measured from one camera (or at most a few averaged). So the 3 by 3 matrix is unlikely to exactly match any given user's camera. Non-standard lightling can also require a different 3 by 3 matrix. So Camera Raw provides a way to tweak this matrix via its controls. A 3 by 3 matrix has 9 degrees of freedom, since it has nine numbers. If you want to tweak 9 degrees of freedom using slider controls, you need nine sliders.

The first degree of freedom is overall scale. This is controlled by the "Exposure" slider. The next two degrees of freedom are the white balance, which are controled by the "Temperature" and "Tint" sliders.

That leaves six degrees of freedom. The usual way these are represented graphically is as a triangle plotted in xy space. The corners of the triangle are the xy coordinates of the red, green and blue primaries of the camera's color space, and define the camera's gamut. What Camera Raw's six hue/sat sliders in the calibration pane are doing is moving the xy coordinates of the triangle corners. <<

[jbrembat]

joofa, I suppose you are making a wrong use of Lindbloom calculator.

As you said, [0.1881852 0.0752741 0.9911085]are the XYZ for a saturated blue in Adobe RGB with D65 white point

For going to ProPhoto D50 :
change the color model to Prophoto and  click to the "XYZ" button.
You get [88 36 241].

You have not to change the white reference to D50, as your XYZ values are computed with D65.

The same values you can get using a CMM.
In ICC profiles the transform to PCS is "adapted" to D50, as PCS white reference is fixed from ICC to D50 so XYZ values computed from a CMM are different, but the final RGB ProPhoto values are the same you get from Lindbloom.

Adobe 1998 blue is contained in Prophoto. End of story.

Jacopo

[joofa]

joofa, I suppose you are making a wrong use of Lindbloom calculator.

As you said, [0.1881852 0.0752741 0.9911085]are the XYZ for a saturated blue in Adobe RGB with D65 white point

For going to ProPhoto D50 :
change the color model to Prophoto and  click to the "XYZ" button.
You get [88 36 241].

You have not to change the white reference to D50, as your XYZ values are computed with D65.

The same values you can get using a CMM.
In ICC profiles the transform to PCS is "adapted" to D50, as PCS white reference is fixed from ICC to D50 so XYZ values computed from a CMM are different, but the final RGB ProPhoto values are the same you get from Lindbloom.

Adobe 1998 blue is contained in Prophoto. End of story.

Jacopo

I thought this would be a settled issue by now.


Please have a look at the attachment to Iliah Borg's first message in this thread where he works the calculation using Bruce Lindbloom's calculator to show that blue required > 255 and so will clip.

Joofa

[digitaldog]

joofa, I suppose you are making a wrong use of Lindbloom calculator.

My suspicions too, and the reason why he might refuse to look at tools he has on his system today that others have used to illustrate the gamut of the two spaces (in 3D no less)!

You have not to change the white reference to D50, as your XYZ values are computed with D65.

Yup, again, its user error. And Mark tried to explain this in post #29, it was ignored. 

In ICC profiles the transform to PCS is "adapted" to D50, as PCS white reference is fixed from ICC to D50 so XYZ values computed from a CMM are different, but the final RGB ProPhoto values are the same you get from Lindbloom.

Adobe 1998 blue is contained in Prophoto. End of story.

Agreed on both points!

This is why the entire discussion is so pointless...

Amen to that.

What would be useful is starting a new post about the raw processing, the assignment of profiles, the assumed color space. For example, we know the ACR engine uses ProPhoto RGB in a linear TRC for processing. But I don’t believe Adobe assumes this is the color space (if we can use that term loosely at this point) as the mapping from the assumed spectral sensitivities of the chip to XYZ to get to ProPhoto.

[sandymc]

What would be useful is starting a new post about the raw processing, the assignment of profiles, the assumed color space. For example, we know the ACR engine uses ProPhoto RGB in a linear TRC for processing. But I don’t believe Adobe assumes this is the color space (if we can use that term loosely at this point) as the mapping from the assumed spectral sensitivities of the chip to XYZ to get to ProPhoto.

Andrew,

The ACR mapping is camera to XYZ D50 referred, then from that to RIMM space (ProPhoto gamma 1). The camera to XYZ D50 conversion includes the color temperature transform - according to Eric Chan (if I understood him correctly anyway), a color temp adjustment in ACR/LR reaches all the back up the processing pipeline to the raw data.

The process is laid out in section 6 of the DNG spec: "Mapping Camera Color Space to CIE XYZ Space"

Regards,

Sandy

[joofa]

My suspicions too, and the reason why he might refuse to look at tools he has on his system today that others have used to illustrate the gamut of the two spaces (in 3D no less)!

Ha ha, Digital Dog you are funny. Why have you chosen to ignore Ilaih Borg's calculation. He is participating in this discussion. Why don't you ask him?

Joofa

[jbrembat]

Please have a look at the attachment to Iliah Borg's first message in this thread where he works the calculation using Bruce Lindbloom's calculator to show that blue required > 255 and so will clip.

This is Iliah's use of the calculatorr:
Set AdobeRGB 1998, gamma=2.2, reference white D65, and enter 0-0-255
Press RGB button
Set ProPhoto RGB, gamma=1.8, reference white D50, and press XYZ button
Now you see B>255.

It is wrong.
After getting XYZ from Adobe1998 (using D65 white reference), you get XYZ D65.
You have not to change the white reference to go to ProPhoto.
If you change the white reference from D65 to D50, the calculator assumes that the XYZ values are D50 not D65.
The result is wrong.

If you want to follows the ICC profile way:

Set AdobeRGB 1998, gamma=2.2, reference white D50, and enter 0-0-255
Press RGB button
(you get XYZ D50)
Set ProPhoto RGB, gamma=1.8, reference white unchanged (D50), and press XYZ button

You get same values for ProPhoto [88 36 241].

Jacopo

[Iliah]

It is not wrong, I forced "assign", which explains what is happening.

[Peter_DL]

If you want to follows the ICC profile way:

Set AdobeRGB 1998, gamma=2.2, reference white D50, and enter 0-0-255
Press RGB button
(you get XYZ D50)
Set ProPhoto RGB, gamma=1.8, reference white unchanged (D50), and press XYZ button
You get same values for ProPhoto [88 36 241].

Results are different when Adaptation is set to None,
which is what AbsCol does, as opposed to RelCol - afaik.

Peter

--

[digitaldog]

Why have you chosen to ignore Ilaih Borg's calculation.

Why are you ignoring at least three ways to view the gamut of the two, in 3D based on at least two software products on that Mac you say you have? The third product is simply a download away.

[jbrembat]

It is not wrong, I forced "assign", which explains what is happening.

What do you mean?

It is wrong as you said to calculator that the XYZ values are D50 and they are D65.

Results are different when Adaptation is set to None, which is what AbsCol does, as opposed to RelCol

Of course, but if you use asbsolute you are going to a destination color space with the same white of the source color space.

This is not the way for verifying if a source color is in the destination gamut.

Jacopo

[MarkM]

It is not wrong, I forced "assign", which explains what is happening.

Yes. Try doing it with 50% grey and you will see that you introduce a nasty color cast. Keeping the ref. white at D65 gives you [0.428622, 0.428622, 0.428622]. Grey stays grey which is what one should expect between these spaces. It also agrees with Photoshop.

This is not the same as converting vs. assigning a profile. If you wanted to just assign ProPhoto, you would just keep the original rgb numbers the same. (Makes the math pretty simple). In the case of assigning you would still expect neutral grey to map to neutral grey in two well behaved working spaces.

which is what AbsCol does, as opposed to RelCol - afaik

I'll be the first to admit that I have trouble parsing the ICC specs especially concerning V.2 rendering intent, but I'm pretty sure you need to account for chromatic adaptation when moving into the connecting space regardless of rendering intent. Once there you can choose to scale to your media white point (which also needs to be in D50 by now) or not depending on you intent. Of course I'm prepared to be wrong on this point if there's someone here with a good understanding of calculating rendering intent.

[Edit: the version 4 spec is more explicit:
"6.2   Rendering intents: The colorimetric rendering intents operate on measurement-based colorimetric values as chromatically adapted to the PCS illuminant D50."]

[digitaldog]

Seems to me, there are two processes here for analyzing and viewing the gamut of the two spaces. One group is using a toolset on Bruce Lindbloom’s site to produce values they say prove that colors in Adobe RGB (1998) exceed ProPhoto. But there are a lot of options to select in the calculator, arguments (or misunderstandings?) about what options to pick etc. I really think those that feel this tool provies their point contact Bruce and have him settle this method of analysis. Clearly, one can select (or not select) something on his site to make the values “prove” their point. Too much wiggle room here!

Then on Bruce’s site, he has his own 3D Gamut mapper. Now at least one person here is unable, or unwilling to use the browser that his OS installed to view the two gamuts in question. That alone provides clues to whether or not this discussion is at all worthwhile! Others here have similar tools that agree with Bruce's own grapher in terms of the gamut of the two spaces in question. The proponents of the “Adobe RGB (1998) has colors outside ProPhoto gamut” would serve their arguments better by telling us why at the very least, 4 different such tools that simply require one to specify two color spaces (most often based on existing ICC profiles we use), doesn’t agree with their ideas about Adobe RGB (1998)’s gamut limitation compared to ProPhoto RGB. No need to mess with rendering intents, white point adaptation, these are tools for mapping color space gamut we’ve had for years. Are they wrong? Does the one calculator with all the options prove they are wrong? Or that the gamut plots shown are wrong based on some settings (but not all) of a web based color calculator of which only the designer (Bruce) can confirm or deny.

This doesn’t have to be rocket (or color) science. Its really easy to plot the gamuts of the two spaces 3 dimensionally.

[MarkM]

This doesn’t have to be rocket (or color) science. Its really easy to plot the gamuts of the two spaces 3 dimensionally.

Amen. And for what it's worth, the color science doesn't have to be rocket science either.

[bjanes]

Andrew,

The ACR mapping is camera to XYZ D50 referred, then from that to RIMM space (ProPhoto gamma 1). The camera to XYZ D50 conversion includes the color temperature transform - according to Eric Chan (if I understood him correctly anyway), a color temp adjustment in ACR/LR reaches all the back up the processing pipeline to the raw data.

The process is laid out in section 6 of the DNG spec: "Mapping Camera Color Space to CIE XYZ Space"

Regards,

Sandy

I'm not a color scientist, but will bring up the following for discussion:

As I recall from a previous post by Thomas Knoll on one of the Adobe forums, XYZ does not have a white point. The original color matching experiments were done with emissive monochromatic sources that were roughly red, green, and blue projected on a white screen. The actual wavelengths were chosen for convenience. According to Wikipedia, " The primaries with wavelengths 546.1 nm and 435.8 nm were chosen because they are easily reproducible monochromatic lines of a mercury vapor discharge. The 700 nm wavelength, which in 1931 was difficult to reproduce as a monochromatic beam, was chosen because the eye's perception of color is rather unchanging at this wavelength." The subjects in the experiments matched a given color that was projected on by screen by adjusting the intensities of the three sources that were projected on the screen adjacent to the color to be analyzed. Negative values were need to match some colors as explained in the Wikipedia article. The actual XYZ values were derived mathematically. The subjects in the Wright and Guild experiments presumably were adapted to some color of light, but that is not stated in the article. In any event, the XYZ stimulus values can be related to actual wavelengths of light. The primaries for the various color spaces such as AdobeRGB and ProPhotoRGB can be expressed in an absolute manner in terms of the related CIE xyY values. In CIE L*a*b and the various RGB spaces one must include a white point in order to define a color.

D50 is usually used for this purpose. The articles talk about Illuminant E, which is not a spectrum of a black body radiator, but has equal XYZ tri-stimulus values.

With these considerations, what exactly is XYZ D50 or D65?

Regards,

Bill

[joofa]

XYZ does not have a white point.

Hi Bill,

The natural white point in XYZ space is [1,1,1].

With these considerations, what exactly is XYZ D50 or D65?

There is no XYZ (D50) or XYZ (D65). There can be only stuff such as RGB (D65) or RGB (D50), etc. Because, the fundamental rule is not to mess with either the state of Texas, or the XYZ coordinate system. Technically, one could have done, say XYZ (D50), and there is nothing stopping that as there is no special difference between XYZ or any RGB in that both are just coordinate systems, but that would mess up the structure of colorimetry. You need one canonical reference coordinate system. And, XYZ ([1,1,1]) is that.

Sincerely,

Joofa